International Journal of Multiphase Flow 126 (2020) 103037

Contents lists available at ScienceDirect

International Journal of Multiphase Flow

journal homepage: www.elsevier.com/locate/ijmulflow

A study of two-phase annular flow using unsteady numerical

computations
Abhishek Saxena a,b, Horst-Michael Prasser a,∗
a
Laboratory of Nuclear Energy Systems, Department of Mechanical and Process Engineering, ETH Zurich, Sonneggstrasse 3, 8092 Zurich, Switzerland
b
Laboratory of Scientific Computing & Modeling, Department of Nuclear Energy and Safety Research, Paul Scherrer Institute (PSI), 5232 Villigen, Switzerland

a r t i c l e i n f o a b s t r a c t

Article history: Flow behavior of thin liquid films in an annular flow regime is an important element of the thermal
Received 30 October 2018 hydraulics of a BWR – it controls the heat transfer from the fuel rods to the coolant. Despite its rel-
Revised 7 April 2019
evance, the flow behavior of dynamic liquid films is not well understood even in adiabatic conditions.
Accepted 7 May 2019
To investigate it, unsteady numerical simulations with an interface tracking model were performed in
Available online 8 May 2019
a double subchannel geometry with a P/D ratio of 1.325. Different turbulence models, namely the large
Keywords: eddy simulation (LES) and linear/nonlinear eddy viscosity unsteady-RANS (URANS) models, were tested.
Annular flow A novel approach for generating turbulent inlet conditions in a periodic flow domain was developed to
Passive scalar mixing reduce computational efforts. Validation against experimental data revealed shortcomings of the linear
Liquid films eddy-viscosity RANS model in predicting key flow parameters. By capturing the effects of the secondary
CFD flow structures in a subchannel geometry, improved predictions were obtained with a non-linear SST k–ω
Subchannel
(QCR) turbulence model. Time-averaged liquid film thickness (LFT) and tracer distribution obtained with
LES were found to have the best agreement with experimental data. Instantaneous and time-averaged
velocity profiles were analyzed to understand the influence of the gas-liquid interface. Secondary flow
structures in the subchannel gap region were found to enhance the turbulent mixing of the passive scalar
in the liquid film. This finding is relevant towards the prediction of thermal-hydraulic parameters in a
multichannel flow assembly by accounting for inter-channel mixing phenomena.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction a dispersed form either as liquid droplets in the gas core or gas
bubbles in the liquid film (Fig. 1). The transfer of heat through the
Two-phase annular flow regime is established in boiling water liquid film is modulated by the flow properties of the gas-liquid in-
reactor (BWR) subchannels where a high steam quality is present terface through the action of the so-called ripple and disturbance
due to coolant boiling. The avoidance of critical heat flux (CHF) waves, respectively. The waves are understood to play a significant
is paramount in the normal functioning of the reactor, which can role in the heat and mass transfer in annular flow regime, for ex-
lead to dryout of liquid film on the fuel rod surface. Under such ample as shown in the work of Jayanti and Hewitt (1997). Hence, it
flow conditions, the temperature of the fuel rod may increase sub- is important to model this characteristic unsteady behavior of the
stantially due to the reduced heat transfer coefficient in a steam- flow.
only environment. This can lead to fuel-cladding oxidation and an Large velocities and high Reynolds numbers characterize the
eventual failure of its mechanical integrity. It is important, from an annular flow regime, which makes it highly challenging to simu-
engineering standpoint, to acquire a deeper understanding of the late. Hence, only a few numerical studies have been undertaken in
flow to achieve better fuel utilization in terms of thermal margins this field. A lack of highly resolved experimental data for model
while preventing the occurrence of dryout on the fuel rods. validation is also another reason. Fukano and Inatomi (2003) per-
The annular flow regime consists of a thin liquid film on the formed DNS simulations of horizontal annular flow in a pipe to
wall characterized by a dynamic interface and a gas core in the study the influence of disturbance waves on the liquid film spread-
center of the flow channel. Each phase may also be transported in ing in the circumferential direction. In their DNS study, the grid
resolution employed was much coarser than that used in the cur-
∗
rent study for similar superficial velocities of the two-phases and,
Corresponding author at: ETH Zurich, ML K13, Sonneggstrasse 3, 8092 Zurich,
Switzerland. therefore, not all physical phenomena might have been accurately
E-mail addresses: abhishek.saxena@psi.ch (A. Saxena), prasser@lke.mavt.ethz.ch captured. Rodriguez (2009) performed a DNS study of annular flow
(H.-M. Prasser).

https://doi.org/10.1016/j.ijmultiphaseflow.2019.05.003
0301-9322/© 2019 Elsevier Ltd. All rights reserved.
2 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037

Nomenclature

k turbulent kinetic energy

Dh hydraulic diameter
uτ shear velocity
μt turbulent viscosity
un normal velocity
τw wall shear
ρ density
y distance from wall
i phase volume fraction
Jφ phase superficial velocity
m˙ φ phasic mass flow rate
y+ non-dimensional wall distance
u+ non-dimensional streamwise velocity

to aid in the development of physics-based closure models used in

a multi-field modeling approach. The work was able to capture the
major mechanisms observed in annular flow but was restricted to
an angular section of a pipe.
Fig. 1. Schematic representation of two-phase annular flow regime.
Anglart and Caraghiaur (2011) reviewed an Euler–Euler and
Euler–Lagrange approach to model boiling annular-mist flow
while Liu et al. (2011) simulated roll waves in annular flow. Table 1
Saxena et al. (2016) simulated annular flow in a tight lattice sub- Main flow parameters.
channel geometry using VOF and got a good agreement with neu- D (mm) P (mm) Dh (mm) Jg (m/s) Jl (m/s)
tron tomography data. Recently, Sato and Niceno (2017) performed
20 26.5 18.85 30 0.1
an LES of upward annular boiling flow on a flat surface with an
interface tracking method to study the properties of disturbance
waves. The work of Carlucci et al. (2004) summarizes the correla-
tions and generalized relationships for single and two-phase tur-
the flow geometry is shown in Fig. 2. The liquid was injected
bulent mixing across both vertical and horizontal interconnected
through a special slit, 0.5 mm in thickness, directly on the mock
flow channels. Highly resolved CFD work on passive scalar mixing
fuel-rods. This facilitated in orienting the flow around the rods
in two-phase annular flow is still lacking to date.
resulting in a faster flow development. It also helped define the
The primary objective of this work is to test the performance
boundary conditions conveniently on the inlet of the simulation
of numerical models available in commercial two-phase CFD codes
domain. Flow properties of the liquid film were studied by in-
by validating the liquid film distribution on a half-rod placed in
jecting a highly-conductive passive scalar. It was tracked using a
a double subchannel geometry and tracer mixing behavior within
liquid film sensor (LFS) which is able to detect changes in the
this film. The presence of a gap between the two flow channels
conductance of the medium in contact with it (Damsohn and
allows for the study of flow phenomenon responsible for mass
Prasser, 2009). With a measurement grid of 64 × 16 points around
and momentum exchange between them. The simulations were
a half-fuel rod, the sensor is able to provide a two-dimensional
performed with the code STAR-CCM+ v. 11.06. An experimental
mapping of the conductance distribution with a frequency as high
database of liquid film flow from Damsohn and Prasser (2010), and
as 10 KHz. Both time-resolved and time-averaged flow properties
passive scalar mixing from Saxena and Prasser (2016), is used for
were obtained for code validation. More experimental details can
numerical validation. The numerical estimation of the spreading
be found in the work ofSaxena and Prasser (2016). Details of the
rate of the passive scalar in liquid film permits a direct validation
flow geometry and conditions are mentioned in Table 1. The flow
of the model by comparing against experimental tracer data. The
rates were selected to limit liquid entrainment by the turbulent gas
use of unsteady turbulence models, in the form of LES and URANS,
flow thereby focusing on simplified flow dynamics.
has been made to obtain time-resolved flow features, particularly
with regards to investigating the dynamic gas-liquid interface. The
high spatio-temporal resolution of the numerical data also allowed
for a more detailed analysis of the liquid film flow properties,
which was not possible with the experimental setup. While a sin- 2. Numerical method
gle experimental test-case is simulated due to the high computa-
tional resources required, a comparison of different flow parame- The different turbulence modeling strategies for unsteady flow
ters, both time-averaged and time-resolved, is done to discuss the simulations make use of the same basic form of the governing
current standing of common modeling approaches for simulating equations through either a time or space-averaging operation of
complex multiphase flows. the Navier–Stokes equations. Even though LES differs from RANS
Prior to the description of the modeling strategy, a few de- modeling in its ability to capture the effects of large-scale mo-
tails of the experimental setup are provided to help establish the tions in the flow, the small-scale dissipative flow phenomenon
background behind this work. Experimental data was made avail- needs to be modeled. With respect to these turbulence models
able from two-phase annular flow experiments performed at ETH only a few details are mentioned here while the readers are re-
Zurich in an adiabatic setting in a vertical, double-subchannel ge- ferred to the works of Moeng (1984) for theoretical details and
ometry. The channel length of 2.5 m was equivalent to a flow de- Labourasse et al. (2007) for applications of LES in two-phase flows,
velopment length of 70 hydraulic diameters. The cross-section of and of Durbin (2002) for URANS.
 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037 3

Fig. 2. Geometry of computational domain.

2.1. Physics models eddies in the gas-phase are suspected to have a considerable effect
on the mixing process within the liquid film through the transport
2.1.1. Turbulence models by large-scale interfacial structures. Hence, the LES approach is jus-
LES is a turbulence modeling approach based on spatial filtering tified for studying two-phase annular flow in the current context.
of the Navier–Stokes equations. It allows the resolution of larger If, on the other hand, one were to study mass or species-transfer
eddies on the grid level while modeling the effects of smaller ones across the gas-liquid interface, the rate-controlling processes would
that have a more isotropic nature. The LES approach is effective be expected to be largely driven by the small-scale motions. There-
when used for predicting mass, momentum and heat transfer in fore, the accurate resolution of the small-scale flow motion would
flows where the transport processes of interest are dominated by be indispensable in such a case. For such an analysis, LES predic-
the resolved, large-scale motions with a subsequent cascade of en- tions have a first-order dependence on these models. RANS mod-
ergy to the statistically isotropic and universal small scales. On the els, on the other hand, are better suited to modeling unsteadiness
contrary, in applications where the controlling processes occur at in the flow due to geometric heterogeneity rather than in station-
these smallest scales, such an approach might be less appropriate. ary flow.
For near-wall flows at high Reynolds numbers, an important In channels of non-uniform flow cross-sectional area, sec-
phenomenon is the momentum transfer due to the interaction of ondary flow structures of Prandtl’s second kind may develop
the flow of the outer layers with the boundary layer flow. While (Bradshaw, 1987). Linear eddy viscosity models assume isotropy
these motions can be resolved by LES, the flow in the region close for the Reynolds stress components, an assumption which may not
to the wall is comprised of small length scale viscosity-dominated hold true for flows with varying boundary layer thickness. Work by
structures. In order to avoid huge computational costs required to Ikeno and Kajishima (2010) has shown that the turbulence energy
resolve this near wall region, models must be employed to cap- transfer in the rod gap is suppressed in the wall-normal direction
ture the relevant flow phenomena close to such boundaries. By and released in the wall-tangential direction. Models that take into
virtue of the large density differences between the gas and liq- account the anisotropy between the normal turbulent-stress com-
uids in annular flow in practical applications, the gas flow near ponents are known to help better predict flows in which secondary
the interface may be seen as analogous to turbulent flow close flow structures exist. See for example Benhamadouche (2017),
to a wall. Since the small-scale motions in the gas flow may be Vitillo et al. (2015), and Baglietto (2006). To account for anisotropic
coupled to and associated with localized small-scale interfacial effects, a non-linear SST k–ω turbulence model with a quadratic
structures, they are not considered to be influential in determin- constitutive relation (QCR) between the strain and vorticity-rate
ing global time-averaged liquid film thickness (LFT) distribution or tensors is tested and compared against a two-layer realizable k–
tracer mixing within the liquid film phase. That is, only large-scale ε model with a linear eddy-viscosity correlation. The relation of
4 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037

Table 2
Comparison of flow parameters.

Temp (°C) Pressure Rod diameter (D) Hydraulic diameter Mass flux Outlet steam
(MPa) (mm) (Dh ) (mm) (kg/m2 s) quality/gas mass
fraction

BWR (in operation) ∼285 7.6 ∼12.3 ∼15 (central subch.) ∼30 0 0 ∼0.15
Experiments (Damsohn and ∼25 0.1 20 18.85 ∼40–235 ∼0.15–0.54
Prasser, 2010; Saxena et al., 2016)

Table 3
Details of computational mesh.

Cell count z (mm) θ avg (o ) rmin (mm) t (μs)

Realizable k–ε 14,980,0 0 0 0.237 1.098 0.006 20
SST k–ω, QCR 14,980,0 0 0 0.237 1.098 0.006 20
LES 28,10 0,0 0 0 0.237 0.720 0.005 5

the Reynolds stress tensor, T, to the strain rate tensor, S, and the along with the LES model in STAR-CCM+. The ratio is formulated
vorticity tensor, W, when using the QCR non-linear eddy viscosity as
model is given in STARCCM+ as follows: ksgs
M= (2)
T = 0.0929 × μt (S.W − W .S ) ksgs + kres
where ksgs is the SGS turbulent kinetic energy and kres is the re-
To model the effects of the subgrid scale turbulence, the Wall
solved turbulent kinetic energy. The resolved turbulent kinetic en-
Adapting Local-Eddy (WALE) model (Nicoud and Ducros, 1999) was
ergy in the LES can be estimated by:
used in the framework of LES. A default value of 0.9 was used as
the turbulent Schmidt number with all the turbulent models. 1
kres = (u˜i − u˜i T )(u˜i − u˜i T ) (3)
2
where u˜i T is the time averaging operator over the velocity field
2.1.2. Interface-tracking model
given by u˜i . A value close to zero indicates a refinement appro-
The VOF model (Hirt and Nichols, 1981) is an interface-tracking
priate for DNS studies whereas a value of one is acceptable only
method in which a single velocity field represents the multiple
for RANS modeling. The instantaneous and time-averaged values of
phases while the volume fraction field is solved using a convection
this ratio are shown in Fig. 4. Although some of the instantaneous
equation:
values close to the interface were over 0.15 on the gas-phase side,
∂ i the time-averaged values were well below 0.1, confirming the qual-
+ u.∇ i = 0 (1)
∂t ity of the grid.
where u is the flow velocity field. The volume fraction field is then
2.3. Inlet conditions
geometrically reconstructed to obtain the interface between the
various phases.
The generation of fully turbulent inlet conditions is necessary
The default HRIC scheme (Muzaferija, 1998), which provides a
for performing unsteady simulations. The superposition of pseudo-
smooth blending of HRIC in low CFL (Courant–Friedrichs–Lewy) re-
random coherent fluid motions to mean values is important in ob-
gions and an upwind scheme in high CFL regions of the flow, was
taining the correct growth and sustenance of turbulence in the
used for this work. The final scheme includes factors for interface
computational domain for an LES study. This is borne out of the
sharpening applicable in the case of excessive numerical diffusion.
fact that the grid resolved structures always include a time-varying
A good result was achieved when it was increased to 0.1 for LES.
component, which need to be taken into account at the inlet by
The interface was identified as the iso-surface of the liquid volume
some mechanism to generate stochastic fluctuations that approxi-
fraction variable with a value of 0.5. A small time-step (in Table 2)
mate ‘turbulence’. Allowing the flow to develop spatially in a chan-
helped to limit CFL < 1 close to the interface.
nel spanning several hundred hydraulic diameters in length is too
computationally expensive in most cases. As the experimental data
2.2. Mesh was obtained under developed flow conditions, a periodic compu-
tational flow domain was deemed necessary to have a valid one-
A high density mesh close to the walls was generated to resolve to-one comparison.
the dynamics of the liquid film. To achieve this, a structured mesh Several techniques can be used to generate inflow turbulence
with fine prism layers around the fuel rod surfaces was employed. for LES studies. Keating et al. (2004) have classified these tech-
Horváth and Dressel (2012) have previously tested the influence of niques into three types, i.e., the recycling methods, the precur-
the structure of the mesh on single-phase flow in a square-lattice sor databases, and the synthetic turbulence methods. For the for-
bundle geometry and found that better results are obtained with mer two kinds of generation methods, auxiliary computational do-
structured, body-fitted grids. The mesh distribution across a cross- mains, which are isolated or connected with the main domain,
sectional plane of the geometry used for the LES is shown in Fig. 3 are needed. As these methods utilize the same flow geometry as
while the cell sizes are listed in Table 3. The minimum cell thick- the main case, the turbulent structures develop naturally inside
ness in the radial direction is found at the wall. the flow domain itself. The synthetic turbulence methods work on
A measure of the grid quality for an LES study is the ratio the principle of superposing random fluctuations with given mo-
of the sub-grid scale turbulent kinetic energy to the total turbu- ments and spectra on to the mean velocity profile. Their develop-
lent kinetic energy. This parameter can be estimated through the ment is still an active field of research (Smirnov et al., 2001; Bat-
SGS turbulent kinetic energy function (SGSTurbulentKineticEnergy) ten et al., 2004). The synthetic eddy method (SEM), proposed by
which is made available by turning on the aeroacoustics model Jarrin et al. (2006), is the in-built method in STAR-CCM+.
 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037 5

Fig. 3. Computational mesh used for LES.

Fig. 4. Time-averaged (left) and instantaneous (right) distribution of M ratio.

The precursor simulation method makes use of a separate quirements. Also, periodic recycling conditions are also employed
initial simulation of the flow to generate turbulent inlet con- for precursor simulations. Since such a database is available only
ditions. The result of the precursor simulations is stored to be for a finite time-period, it may introduce periodicity into the tur-
used later for the main computation. The generation of precur- bulent inlet data leading to an unrealistic periodicity in the flow
sor databases is often cumbersome due to the large memory re- itself.
6 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037

Fig. 5. Periodic flow domain for numerical computation.

Fig. 6. Characteristic length scale in flow direction.

Fig. 7. Instantaneous streamwise velocity field with the recycling method (LES).

2.4. The recycling flow method convergence. A slightly different method of tackling the problem is
by defining a mean pressure gradient in the streamwise direction.
The simplest method for generating turbulent inlet conditions This allows the flow to develop with the phasic mass flow rates
is based on recycling the data from the outlet to the inlet of the associated with the applied mean pressure gradient. If this mean
calculation domain (see Fig. 5). A direct mapping of the flow field pressure gradient is not known a priori, an iterating procedure to
from the outlet to the inlet can result in a slightly different mass estimate it is then required to recover the phasic mass flow rates.
flux to be carried over due to error propagation arising from solver To avoid this tedious procedure, a method based on the internal
 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037 7

The mass flow rate of each phase across any particular flow cross-
section varies with time due to the presence of unsteady flow fea-
tures like waves, etc. Hence, the given equality doesn’t hold any
longer:

m˙ φ = ρφ .Aφ .unφ = const

As a result, the same rescaling method cannot be applied for a
two-phase flow using the instantaneous velocity field at the out-
flow plane as in Eq. (4). Instead, the phasic mass flow rates aver-
aged over the time-period of the longest wavelength can be con-
sidered constant. Since the use of a time-averaging operator is not
suitable for providing time-resolved turbulent inlet conditions, the
space-averaging operator may instead be used. In essence, one can
substitute the time averaging operation over the time-period of the
longest wavelength at a certain cross-section with a spatial aver-
aging operation of the flow over the entire wavelength at some
particular instant of time. In mathematical terms, this is described
Fig. 8. Asynchronous LFT time series, experiments vs. CFD (θ = 90° ). by the ergodic assumption, which states the equivalence of spatial
and time averages of a stationary homogeneous turbulent flow.
 t+T  x+λmax
mapping method of Baba-Ahmadi and Tabor (2009) has been de- 1 1
Q (t ∗ )dt ∗ ≡ Q (x∗ )dx∗
veloped for two-phase flows as part of this work. T t λmax x
The method requires the knowledge of only the mass flow rates The choice of the averaging length λmax is correlated to the in-
of the individual phases. Before outlining the method developed tegral length scale of the computational domain (given by 2.π .Dh
for two-phase flows, the existing method for single-phase flow is where Dh is the hydraulic diameter) or the largest wavelength
described. The mass conservation principle implies a constant mass (Fig. 6).
flow rate for an incompressible single-phase flow through a con- For a given superficial velocity Jφ for phase φ and channel
duit: cross-sectional area Ach , the integral phasic mass flow rate in the
m˙ o = ρ .Ach .un = const domain is given by:
 V
The area of the channel Ach being constant, the outlet velocity m˙ φo dV = ρφ · Jφ · Ach · λmax
field can then be mapped back to the inlet through a scaling factor 0
 V 
m˙ φ dV = ρφ · [(Acell · hcell · un ) · i ] (5)
uin = (m˙ o/m˙ out ) · uout (4) 0 V

where uin is the mapped inlet velocity field using the outlet ve- where Acell · hcell is the volume of the cell, un the streamwise veloc-
locity uout from the previous time-step. Only velocities normal to ity component, i the phasic volume fraction and the summation
the inlet and outlet are used in this step. The rescaling factor lim- of the bracketed quantity over the complete volume V of the do-
its and corrects the propagation of numerical errors in the time main. The quantity m˙ φo represents the target phasic mass flow rate
evolution of the flow to maintain a constant mass flow rate m˙ o. and can be determined prior to the simulation from the density
For two-phase flow, both the velocity field and the phasic volume and superficial velocity of each phase, and the geometric charac-
fraction variables are required to be recycled. The behavior of two- teristics of the flow domain, namely the cross-sectional area and
phase flow is different from single-phase flow in another aspect. the axial flow length. The mapped inlet velocity from the outlet

Fig. 9. (a) LFT probability distribution and (b) RMS of LFT at point probe (θ = 90° ).
8 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037

Fig. 10. Instantaneous LFT distribution, Experiments vs. Two-phase CFD (64 × 16 grid).

3. Results and discussion

The flow was initially allowed to develop for 10 flow through

times based on the mean liquid film velocity given at the inlet
(1.2 m/s). The necessary condition for flow development was deter-
mined by the asymptotic behavior of the turbulent kinetic energy
(TKE) monitor averaged over the flow domain. Flow statistics were
collected over an additional 10 flow through times.

3.1. Liquid film characteristics

The distribution of the liquid film over a fuel rod is discussed

in this section. Several researchers, e.g. Chien and Ibele (1964) and
Fig. 11. Time-averaged LFT around half rod.
Ueda and Nose (1974), have found experimental evidence of a two-
layer flow structure in liquid films. This flow structure has been
widely utilized in the modeling of annular flow since its proposal,
for example by Dobran (1983). The continuous layer close to the
plane then becomes: wall is estimated to be relatively thin and associated with a low
heat transfer coefficient. A wavy layer, which is intermittent in na-
 V  V
 ture, is present on top of this base layer. The waves are known to
uφin = m˙ φo dV m˙ φ dV · uφout (6) travel at much higher velocities, carrying a significant part of the
0 0
liquid mass flow in the process. The regions under the disturbance
waves have been experimentally found to have preferential nucle-
The integrals in Eq. (6) can be easily calculated in any com- ate boiling sites by Barbosa et al. (2002).
mercial code, for example using field functions and reports in An asynchronous LFT time series extracted at the subchannel
STAR-CCM+. The developed velocity field, obtained through the gap position (θ = 90°) is shown in Fig. 8. The dynamics of the
natural development of the turbulent structures with the recy- wavy layer were found to be dependent on the turbulence model
cling method, is shown in Fig. 7. The high turbulent intensities used. Probability density functions (PDFs) of LFT comparing differ-
and strain rates may cause the SEM to generate a large number ent turbulence models are shown in Fig. 9(a). To measure the in-
of small eddies with high vorticities close to the gas-liquid inter- tensity of LFT fluctuations, the RMS of the film thickness was es-
face. Through the recycling flow method, concurrent distributions timated at the subchannel gap position (90°). The LES predicted a
of the velocity field and phase distributions are simultaneously high-peak low-variance distribution around the mean film thick-
made available at the inlet, which is not possible with the SEM ness value due to the absence of any large-amplitude wavy struc-
method. An initial film of 500 μm thickness was set as the starting tures. The realizable k–ε model predicted a thicker wavy layer
condition on all the rods, the rest being gas in the channel core. compared to the experiments with a peak around 700 μm. The
The phasic velocities were then estimated by the given superficial largest RMS of the film thickness was obtained with the non-linear
velocities and the flow area occupied by the two phases. k–ω model (see Fig. 9(b)).
 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037 9

Fig. 12. Secondary flow structures from mean flow.

Fig. 13. Tangential velocity vectors in subchannel gap region (LES).

Snapshots of the unsteady gas-liquid interface obtained with cumference (Damsohn and Prasser, 2010). The blockage of the side
the different turbulence models around a half fuel rod, resolved on subchannel gaps is not deemed to be the cause of this observation
a 64 × 16 mesh, are showed in Fig. 10 and 12. Disturbance waves due to the symmetry about the central subchannel gap. Although
may exist in a stable state for several wavelengths in the flow di- film structures with high LFTs were predicted by the realizable k–
rection. These were distinguishable in the experiments as packets ε model, they were observed to have a low circumferential coher-
of thick film and recorded to be asymmetric about the fuel rod cir- ence. The SST k–ω non-linear eddy viscosity model predicted tilted
10 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037

Fig. 14. TKE distribution from two different URANS models.

Fig. 15. Variance of cross-velocity (uj ) from LES.

or oblique wavy structures, similar to those observed in the exper- two grid points next to the corners. A good agreement was found
iments. The sharpening coefficient used for the VOF discretization between LES, the SST k–ω model and the experiments. Although
scheme with the LES model may prevent the generation of coher- the realizable k–ε model predicted the mean LFT to an accuracy
ent roll waves from localized ripple waves, leading to the direct of 2%, the spanwise distribution revealed a peaked distribution in
dispersion of the liquid phase as droplets in the gas phase instead. the subchannel gap region with a large deviation from the experi-
Moreover, some two-phase turbulence effects may become signif- mental distribution. The non-linear SST k–ω model gave good pre-
icant on approaching the interface due to the large density gradi- dictions for the mean LFT far away from the rod ends and repro-
ent. For example, a model to account for turbulence damping near duced the peaks observed at around 55° and 125° polar positions
the gas-liquid interface, currently absent in commercial codes with on the rod. The secondary flow structures (Fig. 12) were found to
LES, may help to sustain such waves in the flow. Based on these affect the lateral distribution of the LFT on fuel rods through cross-
results, it is recommended to study these effects on the flow in channel mass transport. The improved predictive performance of
future work. the non-linear eddy viscosity correlation in the SST k–ω model is
Time and space-averaged (in streamwise direction) LFT distribu- linked directly to its ability to resolve the secondary flow struc-
tion profiles across the span of a half fuel rod are shown in Fig. 11. tures.
A quantitative comparison of the performance of different models Inter-channel exchange of momentum and energy by large-scale
is shown in Table 4. The mean LFT was calculated by ignoring the coherent structures is well documented for single-phase flows, for
A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037 11

Fig. 16. Planar-velocity vector field near corners (LES).

Fig. 17. Velocity vectors in gas flow above wave (LES).

Fig. 18. Velocity vectors in liquid wave (LES).

12 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037

Fig. 19. Instantaneous snapshot of streamwise velocity field with highlighted gas-liquid interface (LES).

Table 4
Comparison of mean LFT.

Mean LFT (μm) Error, mean Maximum pointwise θ , Maximum

LFT (%) error, LFT (%) error (o )

Experiment 245 0 0
LES 253 3.3 9.0 74
Realizable k–ε 240 2.0 46.2 85
SST k–ω, QCR 224 8.6 6.7 53

example by Rehme (1992). Their existence in the gas-core of an- computed by LES is shown in Fig. 15. In the given computational
nular two-phase flow and their influence on the liquid film cov- domain, a significant contribution to the cross-channel flow effects
ering the fuel rods remains unknown due to the difficulty in ob- comes from the v component. Peaks in both the TKE and the vari-
taining such evidence. These periodic structures can suppress or ance distribution close to the time-averaged interface on the gas
delay CHF by increasing turbulent mixing and preventing the for- side is due to the dynamic nature of the interface, similar to the
mation of local hot spots on fuel rods. Local eddies in the span- behavior near a wall. Higher variance values were observed close
wise direction were predicted by LES in the subchannel gap re- to the open subchannel gap compared to the subchannel centers
gion (see Fig. 13) although no streamwise coherence was detected as the eddies formed near the subchannel gap are not constrained
between them. Increased turbulent activity was recorded in the by any no-slip boundaries.
subchannel gap by the LES but was absent in the URANS predic- The mass transport of the liquid caused by the corner eddies
tions. It should be noted that for large P/D ratios like in the current in the gas flow (see Fig. 16) results in the large film thickness
work (P/D = 1.325), strong flow pulsations due to coherent turbu- recorded at the polar extremities. As mentioned by Damsohn and
lent structures in a subchannel geometry have not been measured Prasser (2010), such an accumulation of the liquid in the corners
(Meyer, 2010). results in a lower LFT on the rods than in a real bundle for the
Time-averaged TKE distribution comparison between the lin- same flow rates. Although this effect may not be relevant to flow
ear (k–ε ) and non-linear (k–ω) eddy viscosity model is shown in in central subchannels, it is nevertheless an important observation
Fig. 14. The distribution of the variance of the cross-velocity, (v )2 , for corner subchannels that can be reproduced computationally.
 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037 13

Fig. 20. Instantaneous snapshot of liquid fraction field with highlighted gas-liquid interface (LES).

Perturbation of the gas-phase flow by large wavy structures found immediately downstream of the waves indicating reversal of
of the liquid film can result in a coupling of the flow dynamics liquid flow all the way to the wall.
near the interface. Fig. 17 shows gas velocity vectors over a liquid Due to the nature of the measurement apparatus, no experi-
wavy structure in the subchannel gap vertical cross-section. The mental data for the gas-phase was available. The dispersed liquid-
gas flow was found to accelerate on approaching the wave crest phase, captured as elongated spherical droplets in the gas-core
due to flow-area blockage, eventually resulting in flow separation by LES, was found to travel at lower flow velocities compared to
further downstream. Negative interfacial shear-stresses were pre- the surrounding gas phase. These can be seen as distinct circular-
dicted downstream of some waves as a result. The pressure differ- shaped regions in the axial velocity field shown for several cross-
ential resulting from the flow separation over a wave leads to a sections from the LES in Figs. 19 and 20. In the absence of experi-
movement of the gas from the core-region towards the liquid film mental data, no further details on the dispersed droplets were an-
in a direction perpendicular to the interface. This leads to an in- alyzed.
creased interfacial shear stress on the windward side of the liquid
film resulting in the growth of such wavy structures. That is, the 3.2. Velocity profile of liquid film
asymmetric distribution of shear stresses on the two slopes of a
wave is coupled to its growth and sustenance. This may eventually There exists a broad consensus regarding the flow structure in a
result in the shearing off of liquid from the tips of large waves as turbulent boundary layer for single-phase flows as defined by the
has been discussed by Kataoka et al. (1983). A similar wave forma- log-law proposed by Von Karman (1939), with typical values of the
tion mechanism was also found in the numerical work of Sato and constants κ = 0.4 and C = 5.5. The Universal Velocity Profile (UVP)
Niceno (2017). These results indicate how the gas-phase bound- is formulated as:
ary layer is highly perturbed by the interfacial features and, hence, 
needs to be taken into account when modeling annular flow. y+ 0 < y+ < 5
+
The shear and normal stress balance at the interface allows the u = 5 log y+ − 3.0 5 < y+ < 30 (7)
+
coupling of the gas-liquid flow in that region. Velocity vectors un-
1
κ log y + C y+ > 30
der a liquid wave are shown in Fig. 18. The liquid close to the inter- where
face is accelerated by the higher gas-shear on the windward side √
while the leeward side gets decelerated. Negative velocities were u+ = u/uτ with uτ = τw /ρ y+ = yuτ /ν
14 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037

Fig. 21. Time-averaged velocity profile in liquid film in dimensionless (left) and dimensional (right) units.

Fig. 22. Time-averaged tracer distribution in liquid film around half-rod (64 × 16 grid).

It is not known if the classical log-law behavior is also valid This motivated the investigation of the time-averaged flow
for shear-driven liquid films. Due to the action of turbulent gas structure of liquid films from the LES data. The velocity profiles
shear at the interface, the flow structure of a liquid film in an obtained from this work are plotted in Fig. 21. In the laminar sub-
annular flow is thought to be more complex than that of a layer region, i.e. for y+ ≤ 5, the velocity profile was found to obey
single-phase, turbulent boundary layer. Measurements performed the linear region of the UVP. Above this layer, the velocities found
by Vassallo (1999) confirmed the preservation of the single-phase by LES were larger than those from the UVP but close to those ob-
log-law in thin liquid films. However, for describing thicker liquid tained from the profile of Shedd (2013).
films, a modified log-law with a smaller value of C was proposed. The distinct flow structure of liquid films results from the pres-
It was even proposed to be as low as C ≈ 0.0 for an air-water an- ence of waves that carry a large momentum flux, a feature absent
nular flow with Jl = 1.3 m/s, Jg = 12.8 m/s and a measured film in single-phase flows. The formation of large eddies in a direc-
thickness of 0.69 mm. The velocity profile in the liquid film was tion normal to the interface is damped due to the surface tension
found to flatten on approaching the film-gas interface in the work forces acting along the interface. Hence, this decrease in the tur-
of Vassallo (1999). Liquid film velocity profiles recently obtained bulent mixing length near the interface leads to higher velocities
by Shedd (2013) through PIV were also found to deviate from the in the liquid film as compared to single-phase flows at a given dis-
UVP. A modification of the coefficients in the buffer and the log tance away from the wall. To take this into account, a modified
layer was proposed by the authors as: velocity profile for liquid films is proposed here. Considering the
effects of the two shear layers, namely the wall and the interfacial
 shear layer, on the liquid film, an ‘S’-shaped velocity profile is pro-
y+ 0 < y+ ≤ 5
+ posed here. Since a logarithmic profile is found to fit the so-called
u = 7.2 log y+ − 6.6 5 < y+ ≤ 30 (8)
buffer layer near the wall, a power profile of the form a(y+ )e + b is
7.38 log y+ − 7.1 y+ > 30
 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037 15

in the velocity profile of the liquid film, which the proposed pro-
file satisfies. Since only one set of flow rates are simulated in this
work, the dependency of the coefficients a and b on the superfi-
cial velocities and fluid properties remains to be studied in future
work.

3.3. Tracer mixing

Two-dimensional, time-averaged tracer distributions from the

experiments and computations are shown in Fig. 22. The tracer
distribution is similar to that characteristic of jets or plumes in
terms of the lateral spreading downstream of the injection point.
The cross-transport of the tracer is a result of the action of the
transverse gas eddies on the liquid film. Time-averaged width of
the plume extracted from the experiments and numerical simu-
lations is plotted in Fig. 23. The width was obtained by fitting
Gaussian curves over data points on each row of the matrix and
is equivalent to the full width half maximum (FWHM) multiplied
by a constant (∼1.18). The limited time statistics of LES data, due
to the large computational requirements, result in a slightly non-
Fig. 23. Tracer mixing width downstream of injection.
smooth distribution comparatively.
A good agreement for the mixing width was obtained between
chosen to characterize the velocity in the top wavy layer. The fol- the experiments and LES, whereas the realizable k–ε model signif-
lowing fit for the streamwise velocity profile from the LES data is icantly under predicted it. The SST k–ω model predicted the mix-
obtained as: ing width well till 30 mm downstream of the injection point but
 √ to deviate. Interestingly, the SST k–ω model predicts
then starts
y+ 0 < y+ ≤ 5
a classic d behavior of the mixing width as a function of dis-
+
u = 7.54 log y+ − 8.276 5 < y+ ≤ 25 (9)
tance. Eddies in the gas flow have a direct effect on the formation
8.923 × 10−4 (y+ )2.718 + 10.89 y+ > 25
of interfacial structures in the liquid film and consequently on the
It is to be noted that the profiles defined here characterize the mixing of passive scalar in the liquid film. The under-prediction of
dependency of the velocity on only the wall distance and no ef- the mixing by the k–ε model is supported by the presence of the
fect of the liquid holdup is included. In other terms, the integra- peak in the LFT distribution at the subchannel gap (Fig. 11). Even
tion of the proposed velocity profile over the film height will fail though the non-linear viscosity turbulence model is able to repro-
to provide accurate mass flow rates for the liquid phase. Instanta- duce secondary flow in the channel geometry, it is still unable to
neous velocity profiles similar to the ‘S’-shaped profile shown in capture the cross-channel eddies that sustain and enhance mixing
Fig. 21 have been previously reported by Lorencez et al. (1997) in far away downstream of the injection point of the tracer. Since the
liquid stratified flows. Evidence of eddies under waves has already LES approach resolves the inertial subrange scales of the flow, it is
been mentioned (Hewitt and Jayanti, 1997) and shown in this work able to better predict this cross-channel mixing compared to the
(Fig. 18). By Fjorth’s theorem this would require an inflection point URANS models.

Fig. 24. (a) Instantaneous transverse velocity and (b) variance of transverse velocity and (c) transverse tracer flux in liquid film (LES).
16 A. Saxena and H.-M. Prasser / International Journal of Multiphase Flow 126 (2020) 103037

Large fluctuations of the transverse velocity of the liquid film Bradshaw, P., 1987. Turbulent secondary flows. Ann. Rev. Fluid Mech. 19 (1), 53–74.
(averaged over film thickness) were recorded at the subchannel Carlucci, L.N., Hammouda, N., Rowe, D.S., 2004. Two-phase turbulent mixing and
buoyancy drift in rod bundles. Nucl. Eng. Des. 227 (1), 65–84.
gap location (see Fig. 24(b)). These fluctuations are responsible Chien, S.-F., Ibele, W., 1964. Pressure drop and liquid film thickness of two-phase
for the enhanced transport of the tracer in the azimuthal direc- annular and annular-mist flows. J. Heat Transf. 86, 89–96.
tion across the rod. This finding is consistent with the high tur- Damsohn, M., Prasser, H.-M., 2010. Experimental studies of the effect of functional
spacers to annular flow in subchannels of a BWR fuel element. Nucl. Eng. Des.
bulent activity observed in the subchannel gap region. The cross- 240 (October (10)).
movement of the gas-phase, through the action of shear stresses, Damsohn, M., Prasser, H.-M., 2009. High-speed liquid film sensor for two-phase
causes fluctuations in the azimuthal mass transport of the film lo- flows with high spatial resolution based on electrical conductance. Flow Meas.
Instrum. 20 (March (1)).
cally. The absence of these large eddies in URANS results in an
Dobran, F., 1983. Hydrodynamic and heat transfer analysis of two-phase annular
under-prediction of the passive scalar mixing. Hence, a thermal- flow with a new liquid film model of turbulence. Int. J. Heat Mass Transf. 26,
hydraulic analysis of annular flow in a rod bundle geometry should 1159–1171.
Durbin, P.A., 2002. A perspective on recent developments in RANS modeling. In:
take into account turbulent mixing effects across subchannels.
Engineering Turbulence Modelling and Experiments, 5, pp. 3–16.
Fukano, T., Inatomi, T., 2003. Numerical analysis of liquid film formation in a
4. Conclusions horizontal annular flow. Int. J. Multiphase Flow 29, 1413–1430. doi:10.1016/
S0301-9322(03)00127-7.
Hirt, C.W., Nichols, B.D., 1981. Volume of fluid (VOF) method for the dynamics of
Unsteady two-phase CFD simulations of the annular flow free boundaries. J. Comput. Phys. 39, 201–225.
regime were performed and validated against both phase- Horváth, A., Dressel, B., 2012. Numerical simulations of square arrayed rod bundles.
distribution and tracer mixing experiments in a geometry relevant Nucl. Eng. Des. 247, 168–182. doi:10.1016/j.nucengdes.2012.03.004.
Ikeno, T., Kajishima, T., 2010. Analysis of dynamical flow structure in a square ar-
to industrial applications. The numerical analysis was made possi- rayed rod bundle. Nucl. Eng. Des. 240 (2), 305–312.
ble by simulating flow in a periodic domain by the development of Jarrin, N., Benhamadouche, S., Laurence, D., Prosser, R., 2006. A synthetic-ed-
a novel method capable of recycling the outflow boundary condi- dy-method for generating inflow conditions for large-eddy simulations. Int. J.
Heat Fluid Flow 27, 585–593.
tions. The LES prediction for the time-averaged film thickness dis- Jayanti, S., Hewitt, G.F., 1997. Hydrodynamics and heat transfer in wavy annular
tribution was in good agreement with the experiments although gas-liquid flow: a computational fluid dynamics study. Int. J. Heat Mass Transf.
some large wavy structures were found to be under-predicted. A 40, 2445–2460.
Kataoka, I., Ishii, M., Mishima, K., 1983. Generation and size distribution of droplet
non-linear eddy viscosity turbulence model, namely the SST k–ω,
in annular two-phase flow. J. Fluids Eng. 105 (2), 230–238.
QCR model, outperformed the linear realizable k–ε model by cap- Keating, A., Piomelli, U., Balaras, E., Kaltenbach, H.J., 2004. A priori and a posteriori
turing the secondary flow effects in the subchannel geometry. The tests of inflow conditions for large-eddy simulation. Phys. Fluids 16, 4696–4712.
Labourasse, E., Lacanette, D., Toutant, A., Lubin, P., Vincent, S., Lebaigue, O.,
liquid film flow structure was found to differ from a single-phase
Sagaut, P., 2007. Towards large eddy simulation of isothermal two-phase flows:
turbulent boundary layer due to the presence of the gas-liquid in- governing equations and a priori tests. Int. J. Multiph. Flow 33 (1), 1–39.
terface. A profile for time-averaged liquid film velocity contain- Liu, Y., Cui, J., Li, W.Z, 2011. A two-phase, two-component model for vertical upward
ing an inflection point is proposed in this work to take into ac- gas-liquid annular flow. Int. J. Heat Fluid Flow 32, 796–804.
Lorencez, C., Nasr-Esfahany, M., Kawaji, M., Ojha, M., 1997. Liquid turbulence struc-
count the two shear layers influencing the flow. Improved pre- ture at a sheared and wavy gas-liquid interface. Int. J. Multiph. Flow 23 (2),
diction of tracer mixing in the liquid film was obtained with LES 205–226.
highlighting the importance of modeling the large eddies in the Meyer, L., 2010. From discovery to recognition of periodic large scale vortices in rod
bundles as source of natural mixing between subchannels—a review. Nucl. Eng.
subchannel gap. Increased turbulent activity in the gap connecting Des. 240 (6), 1575–1588.
two channels, as captured by the LES, has important implications Moeng, C.H., 1984. A large-eddy-simulation model for the study of planetary bound-
with regards to the heat transfer characteristics under annular flow ary-layer turbulence. J. Atmos. Sci. 41 (13), 2052–2062.
Muzaferija, S, 1998. A two-fluid Navier-Stokes solver to simulate water entry. In:
regime conditions in rod bundle geometry. The scope of this work Proc. of the 22nd Symp. on Naval Hydro.
clearly shows the capability of current CFD models in predicting Nicoud, F., Ducros, F., 1999. Subgrid-scale stress modelling based on the square of
the void fraction distribution in BWR subchannels and obtaining the velocity gradient tensor. Flow Turbul. Combust. 62 (3), 183–200.
Rehme, K., 1992. The structure of turbulence in rod bundles and the implications on
liquid film flow properties while other areas have been identified
natural mixing between the subchannels. Int. J. Heat Mass Transf. 35, 567–581.
for further research. Rodriguez, J.M., 2009. Numerical Simulation of Two-Phase Annular Flow. Rensselaer
Polytechnic Institute.
Sato, Y., Niceno, B., 2017. Large eddy simulation of upward co-current annular boil-
Supplementary materials
ing flow using an interface tracking method. Nucl. Eng. Des. 321, 69–81 ISSN
0029-5493.
Supplementary material associated with this article can be Saxena, A., Prasser, H.-M., 2016. Velocity measurements and turbulent dispersion
found, in the online version, at doi:10.1016/j.ijmultiphaseflow.2019. in annular liquid film flows in BWR subchannels with spacers. Nucl. Technol.
dx.doi.org/10.13182/NT16-20.
05.003. Saxena, A., Zboray, R., Prasser, H.-M., 2016. Simulations and measurements of adia-
batic annular flows in triangular, tight lattice nuclear fuel bundle model. Nucl.
References Eng. Des. ISSN 0029-5493.
Shedd, T.A., 2013. Two-phase internal flow: toward a theory of everything. Heat
Anglart, H., Caraghiaur, D., 2011. Cfd modeling of boiling annular-mist flow for dry- Transf. Eng. 34, 420–433.
out investigations. Multiph. Sci. Technol. 23, 223–251. Smirnov, A., Shi, S., Celik, I., 2001. Random flow generation technique for large eddy
Baba-Ahmadi, M.H., Tabor, G., 2009. Inlet conditions for LES using mapping and simulations and particle-dynamics modeling. J. Fluids Eng. 123, 359.
feedback control. Comput. Fluids 38, 1299–1311. Ueda, T., Nose, S., 1974. Studies of liquid film flow in two-phase annular and annu-
Baglietto, E., 2006. Anisotropic turbulence modeling for accurate rod bundle sim- lar-mist flow regions:(part 2, upflow in a vertical tube). Bull. JSME 17, 614–624.
ulations. In: 14th International Conference on Nuclear Engineering. American Vassallo, P., 1999. Near wall structure in vertical air-water annular flows. Int. J. Mul-
Society of Mechanical Engineers, pp. 343–352. tiph. Flow 25, 459–476.
Barbosa, J.R., Hewitt, G.F., Richardson, S.M., 2002. Forced convective boiling of Vitillo, F., Galati, C., Cachon, L., Laroche, E., Millan, P., 2015. An anisotropic
steam-water in a vertical annulus at high qualities. Exp. Therm. Fluid Sci. 26, shear stress transport (ASST) model formulation. Comput. Math. Appl. 70 (9),
65–75. 2238–2251.
Batten, P., Goldberg, U., Chakravarthy, S., 2004. Interfacing statistical turbulence clo- Von Karman, T., 1939. The analogy between fluid friction and heat transfer. Trans.
sures with large-eddy simulation. AIAA J. 42, 485–492. ASME 61, 705–710.
Benhamadouche, S., 2017. On the use of (U) RANS and LES approaches for tur-
bulent incompressible single phase flows in nuclear engineering applications.
Nucl. Eng. Des. 312, 2–11.